Disentangling the Contemporaneous and Life Cycle Effects of Body Mass on Earnings by Donna B. Gilleskie, Euna Han, and Edward C. Norton Donna B. Gilleskie Department of Economics University of North Carolina at Chapel Hill Euna Han College of Pharmacy Yonsei University Edward C. Norton Department of Health Management and Policy Department of Economics University of Michigan December 2015 Preliminary Please do not quote We are grateful for the comments from seminar participants at Elon, Georgia, Georgia State, McGill, NYU, Rochester, Washington University at St. Louis, UNC-Greensboro, Vanderbilt, Virginia Commonwealth, Virginia Tech, William & Mary, Yale, the Annual Health Econometrics Workshop, the Carolina Population Center, and the Triangle Health Economics Workshop.
1 Introduction In this study we quantify the life cycle effect of body mass on earnings of females using data on women followed annually up to twenty years during the time of their lives when average annual weight gain is highest. We allow body mass to explain variation in wages contemporaneously conditional on observed measures of human capital and productivity (namely, education, employment experience, marital status, and family size) and over the life cycle through its impact on the endogenous behaviors that define the time-varying human capital wage determinants. It is common for economists to evaluate the contemporaneous effects of policy and behavioral variables on human capital outcomes such as education, employment, and wages. With increasing interest lately, researchers are studying the effect of early life (and in utero) exposures on these later life outcomes (Heckman, et al., 2013). However, what is lacking in the literature is a dynamic analysis of life cycle impacts of evolving variables on time-varying outcomes. For example, economists may examine the impact of early life health (e.g., birth weight, childhood illness, mother s smoking behavior while pregnant) on later life outcomes (e.g., adult weight, human capital accumulation, mortality, respectively) but, generally, are unable to demonstrate the mechanisms through which variation in early life health affects later life outcomes. This omission often stems from lack of data on individuals at multiple times in their lives over a long period. Importantly, in this example (i.e., health) and in many others, the explanatory variable itself evolves over time as a dynamic process. Likewise, the outcomes of interest (e.g., employment) may vary over time. Hence, as initial health impacts subsequent health, time-varying health may impact the outcome of interest over time. Additionally, causality may run in both directions: for example, health impacts employment and employment impacts subsequent health. Thus, it is important to model the endogeneity of both time-varying variables and to capture the dynamic impacts of each on each other. It has been shown that the body mass of females is negatively correlated with wages (Cawley, 2004). The evidence in much of the literature reflects the contemporaneous relationship between current body mass and current wages. Even in the cases where longitudinal observations on individuals have been available, researchers focus on the impact of current 1
weight status on current earnings, and ignore the evolutionary process of weight gain. 1 While theory suggests avenues through which the history of an individual s body mass may impact wages in the present (e.g., through educational attainment and work experience), the causal relationship between past health (body mass) and accumulation of human capital is rarely included in published analyses of weight on wages. 2 In our effort to understand how wages vary with a determinant of productivity such as body mass that evolves over time, we argue that it is important to consider the effects of this characteristic on other decisions made jointly with employment over the life cycle. Using a long panel of observations on young women as they approach middle age, we capture the influence of an evolving body mass on life cycle decisions pertaining to schooling, employment, marital status, and family size. The history of these decisions, as well as one s body mass history, define cumulative measures of human capital and other productivity measures that explain variation in observed wages. In turn, these behaviors may also impact the evolution of body mass over the life cycle. Even after accounting for the effect of an evolving body mass on the productive determinants of wages that accumulate over the life cycle (e.g., educational attainment, work experience, marital status, and family size), variations in contemporaneous body mass may still explain observed variation in contemporaneous wages. A significant contemporaneous effect may reflect wage penalties associated with lower productivity (i.e., through reduced health) conditional on other observed measures of productivity. That is, the variables that traditionally summarize one s human capital may not fully capture productivity. Alternatively, a contemporaneous effect of body mass may reflect employer preference for employees with a particular outward appearance due to concerns about consumer distaste for obesity or about high expected medical care expenditure affecting firm health insurance costs. Finally, employers may simply taste discriminate against women with higher body mass. In this paper, we establish the existence (or not) of a contemporaneous wage penalty of body mass while also measuring the life cycle impact of body mass on wages; we do not differentiate between individual productivity, employer preference, or employer discrimination as 1 Only a few papers, to our knowledge, consider the effects of current and past body mass on current earnings (Chen, 2012; Pinkston, 2015). Chen includes body mass from a decade prior and Pinkston includes body mass from one year ago. Pinkston s use of lagged wages to explain current wages and twice lagged wages as instruments allows for indirect life cycle body mass effects. 2 Han, et al. (2011) find an indirect wage penalty of body weight in one s late teens through its effect on education and occupation outcomes measured in the late thirties. 2
explanations for the contemporaneous effect. Admittedly, they each may partially explain the current market value (i.e., wage), or marginal product, of an individual s time. Our contribution is the careful, quantifiable distinction between the wage effects of body mass characterized by life cycle human capital differences (i.e., the indirect or life cycle effect) and those reflecting additional differences in current valuations of market work (i.e., the direct or contemporaneous effect). To disentangle these life cycle and contemporaneous effects we use a multiple equation, dynamic empirical model to capture the annual employment decisions and wage distribution of women aged 18 to 45 while jointly explaining education, marital, and family size histories, and the evolution of body mass, over time. Three potential sources of bias compromise measurement of the role of health capital (i.e., body mass) in explaining wages: selection, endogeneity, and measurement error. 3 First, observed wages of employed individuals reflect accepted wage offers, yet we want to measure the labor market s valuation of particular individual wage determinants unconditional on the observed employment decision. That is, individuals who chose to work in the formal labor market are a self-selected group: only those who receive wage offers above an individual reservation wage are observed to work. Estimation using the self-selected sample will result in biased coefficients on explanatory variables for wages if non-random selection into employment is not modeled jointly with the observed wages. 4 Thus, we model non-, part-time, and full-time employment over the life cycle jointly with wages conditional on employment, allowing the observed employment outcome and wages to be correlated through common unobservables. Second, the human capital and other productivity characteristics that determine pay for work reflect life cycle decisions of an individual. These wage determinants include educational attainment, work experience, marital status, number of children, and, as we consider in this paper, body mass. Endogeneity suggests that unobserved characteristics that explain these (time-varying) decisions may also explain differences in wages. 5 Thus, we 3 These same biases impact evaluation of human capital and other endogenous wage determinants, which we also measure. 4 In addition to the effects of body mass on the wage distribution, the probability of receiving an employment offer may be impacted by one s weight, thus influencing accumulation of work experience over the life cycle. Much of the literature on weight and wages ignores this employment selection. 5 For example, a permanent individual trait that may affect both body mass and wages (conditional on body mass) is self-confidence, which is often unobserved in survey data. Although used infrequently in this literature, first-differenced methods and panel data can address endogeneity that may bias the measured weight/wage relationship. Receiving even less attention has been the role of time-varying unobservables that 3
use a flexible, latent factor approach to model permanent and time-varying unobservables that are correlated with wages and their endogenous determinants. Lastly, the measurement error (i.e., reporting, subjectivity, and rounding errors) that accompanies self-reported individual-level characteristics (e.g., weight and height) can lead to inconsistent estimates. 6 Our latent factor method that accounts for individual-level unobservables addresses potential attenuation bias. We use data on women from the National Longitudinal Survey of Youth (NLSY, 1979 cohort). Our research sample includes annual observations from the ages of 18-26 in 1983 through the ages of 37-45 in 2002. These twenty years of data on the same individuals provide a quite comprehensive picture of life changes from young adulthood to middle age. Because these individual behaviors and outcomes (e.g., education, employment, marriage, children, wages, and body mass) are potentially correlated through permanent and time-varying individual unobserved heterogeneity (UH), we use an estimation framework that simultaneously explains variation in the multiple annual events by variation in both observed and unobserved factors. We make few distributional assumptions about the unobservables; rather, we discretize the distributions and estimate the event-specific impacts, mass points, and weights of these unobserved factors using a flexible, discrete factor random effects method. We also seek to understand the impact of body mass on the full support of wages, not just the first moment. That is, we allow body mass (and other variables) to influence wages differently at different points of support of the wage distribution by modeling the density of wages without imposing a distributional assumption. Similarly, we allow determinants of body mass to have different effects at different points of the support of the body mass distribution. We estimate the conditional densities for wages and body mass jointly with endogenous behaviors that determine both wages and body mass, allowing permanent and time-varying observed and unobserved heterogeneity to affect all outcomes. affect both body mass and employment or wages. For example, an unobserved temporary negative health event (e.g., a broken leg) may lead to weight gain as well as a reduction in observed wages if the researcher cannot fully account for accumulation (or depreciation) of human capital (e.g., work absences). 6 There is evidence in the literature that women understate their weight (Lakdawalla and Philipson, 2002; Chen, 2012). Some authors in this literature have re-scaled self-reported height and weight following Cawley (2004); others have used different measures of adiposity (Burkhauser and Cawley, 2008). 4
We extend the literature in several dimensions. First, we analyze the effects of body mass on wages using yearly observations on a cohort of women followed for twenty consecutive years. This point of reference allows for a better understanding of both body mass and wage dynamics and provides a perspective that differs from (repeated) cross-sectional analyses. We are able to quantify the effects of (and determinants of) body mass as an individual ages, and broaden the interpretation from one of aggregate time effects to individual behaviors. Second, we jointly model, along with the evolution of body mass and wages, several individual behaviors that may be affected by body mass and that also explain observed wage variation. Hence, we are able to decompose the correlation between body mass and wages reported in the literature into life cycle effects captured through human capital variables and contemporaneous (direct) influences. Third, we apply dynamic estimation techniques that go beyond traditional methods to better capture impacts of interest over time and across the impacted distribution and to eliminate bias in both the observed and unobserved dimensions. Using our research sample consisting of twenty years of annual observations on the same individuals, we have replicated, with similar specifications and estimation techniques, the results in the literature. The biased findings suggest that current body mass, measured by the conventional body mass index (BMI), has a statistically significant negative effect on the current wages of white women. We find a smaller, but statistically significant effect on the wages of black women in the most sparse specifications. When we take advantage of the longitudinal nature of the data, and estimate a model with fixed individual effects, the upwardly-biased significant effect of BMI on wages of white women falls, but is still significant. The effect for black women in a model that allows for permanent individual unobserved heterogeneity actually reverses signs. We move beyond the standard approaches in the literature to jointly estimate 17 equations that capture the dynamic life cycle behavior described above and that allow for selection into employment, endogeneity of several determinants of wages, and measurement error. Here, calculation of the marginal effect of BMI on wages is not so straight forward because of its contemporaneous and life cycle effects captured by the dynamic model. We quantify those marginal effects through simulation using the estimated model. We find a different effect of BMI at different levels of wages, which is captured by the 5
conditional density estimation procedure. More specifically, the impact of a contemporaneous improvement in body mass on wages of white women is positive and increases over the support of the distribution of wages. That is, the largest impacts occur at the highest deciles of wages. For black women, the contemporaneous weight-related wage penalty is smaller and is actually negative at lower wages, yet approaches that of white women at higher wages. When we account for the impact of better health (i.e., a lower body mass) over the life cycle on other determinants of wages, we find that the total effect on wages is slightly smaller than that implied by the contemporaneous effect. The life cycle impact of body mass (which includes the contemporaneous effect) is significantly different from zero for all women over the support of wages. 2 Review of the Relevant Literature 2.1 Measured Associations between Body Mass and Wages An association between obesity and labor market outcomes, particularly wages, has been documented in the economic literature. The majority of these studies estimate a reducedform static model of the relationship between body mass and wages. A few exceptions explore the underlying mechanisms behind the relationship (Harper, 2000; Baum and Ford, 2004; Bhattacharya and Bundorf, 2004; Carpenter, 2006; Han, Norton, and Stearns, 2009; Han, Norton, and Powell, 2009). Harper (2000) shows a positive effect of being physically attractive on the probability of working in managerial or professional specialty and clerical occupations for women. However, physical attractiveness is not found to be associated with sorting into customer-oriented occupations and also no wage differences are found between non-attractive and attractive women in customer-oriented jobs in the same study. Baum and Ford (2004) examine four potential pathways linking obesity to labor market outcomes: less productivity due to health problems from obesity (measured by health limitations), less investment in human capital by obese workers (measured by work experience), employer distaste for obese employees due to high health care cost (measured by employer provision of health insurance), and consumer distaste for obese workers (measured by classification of 6
occupation as customer-related). Bhattacharya and Bundorf (2004) report no statistically significant wage differential between obese and non-obese individuals by employer provided health insurance coverage. Carpenter (2006) shows that the employment rate increased four percent for obese women and two percent for obese men compared to their respective nonobese counterparts after a 1993 court case (Cook vs. Rhode Island) in which a federal appeals court ruled that obesity is covered under the Rehabilitation Act of 1973 and the Americans with Disabilities Act. Han, Norton, and Stearns (2009) report larger negative relationships between adult contemporaneous body mass and wages in occupations requiring interpersonal skills. Han, Norton, and Powell (2009) examine the direct effect of body mass on wages and the indirect effects operating through education and occupation choice. They find that body weight in the upper tail of the distribution significantly affects educational and occupational outcomes of men and women, which in turn have significant effects on wages. The 11 percent adult wage penalty of obesity during a woman s late teen years can be partly attributed to the effect of body mass on human capital accumulation. While almost all empirical work on this subject estimates ordinary least squares models with or without attempts to address endogeneity of body mass, Johar and Katayama (2012) use quantile regression to assess the role of body mass at different wage levels. They find that the body mass wage penalty is larger at higher wages for white women and at median wages for white males. 2.2 Endogenous Body Mass Only a few of the studies attempt to address the potential endogeneity of body mass, typically measured by BMI or an obesity indicator (Averett and Korenman, 1996; Pagan and Davila, 1997; Behrman and Rosenzweig, 2001; Cawley, 2004; Baum and Ford, 2004; Conley and Glauber, 2005; Norton and Han, 2008; Johar and Katayama, 2012; Sabia and Rees, 2012; and Pinkston, 2015). Instrumental variable estimation techniques require a variable that is correlated with body mass in the current period but uncorrelated with wages in the current period. Instrumental variables (IVs) used in this literature include a lag of respondents own body weight or BMI (Averett and Korenman, 1996; Conley and Glauber, 2005), or siblings or children s BMI (Cawley, 2000; Cawley, 2004; Johar and Karayama, 2012). The lag of current BMI is not a valid instrument if there is any serial or inter-temporal correlation in 7
the wage residuals. Likewise, children s body weight is not a valid instrument if unobserved heterogeneity in the wage residual is associated with both children s body weight and the mother s employment behavior. Also, siblings BMI is not a valid instrument if siblings share unobserved endowment factors that influence earnings. Norton and Han (2008) use genetic variation among young adults to identify their model (using multiple instruments). They find no statistically significant effect of adolescents BMI on wages in young adulthood. There are potential limitations of using own genetic information as IVs for own behavioral outcomes given that genes that predict body weight outcomes are related to brain chemistry and, thus, are likely to affect labor market outcomes such as wages via various channels. Pinkston (2015) estimates the impacts of current and lagged BMI on wages using a differenced GMM estimator which relies on additional lags of body mass as identifying instruments. 2.3 Body Mass and Other Outcomes Regardless of the estimation techniques in the previous economic studies, most studies report gender differences in the association between obesity and wages. Women consistently show a statistically significant negative association of BMI or obesity with hourly wages (Averett and Korenman 1996; Cawley 2004; Baum and Ford 2004; Conley and Glauber 2005; Han, Norton, and Stearns 2009). BMI is also associated with women s total household income by affecting their spouses earnings and occupational prestige (Conley and Glauber, 2005). Some studies also report racial differences in the association between body mass and labor market outcomes for women. For example, Cawley (2004) reports that only white women s hourly wages are causally associated with their body mass. Some studies assess the association of obesity on labor market outcomes at the extensive margin by measuring its effect on the probability of employment or the different effect by different job sectors. Cawley (2000) finds no statistically significant effect of obesity on the amount of paid work and limitations on types of paid work. Morris (2007) estimates a negative relationship between obesity and the probability of employment for British people for both genders. Paraponaris and colleagues (2005) report that a one standard deviation increase of BMI from the mean at age 20 raises the percentage of time spent unemployed 8
during the working years and lowers the probability of employment after a period of unemployment for both genders. Pagan and Davila s (1997) study reports that both obese men and women are less likely to sort into managerial, professional and technical occupations among fourteen Census occupation categories. There are also a few papers estimating the effect of obesity on educational achievement. Some studies report no effect of obesity on grade point average among adolescents (Crosnoe and Muller, 2004), scores on Peabody Individual Achievement Tests among pre-teen children (Kaestner and Grossman, 2009), and grade progression and drop out status (Kaestner, Grossman, and Yarnoff, 2009) among adolescents. In contrast, Sabia (2007) reports that white female adolescents have a grade point average penalty for being obese, whereas non-white female and male adolescents do not. However, Sabia (2007) uses parent-reported self-classification of own obesity as an IV for adolescent BMI, which is not likely to be a valid IV if the parents self-classification reflects the level of their self-esteem or time preference which may also influence their children s educational achievement. The role of body mass on marriage behavior has received much attention from economists. Using a model of the marriage market, Chiappori, et al. (2012) estimate the trade-offs (in terms of income and education) that men and women are willing to accept to compensate for body mass penalties. Malcolm and Kaya (2014) argue that an observed negative correlation between BMI and marriage is not the only consequence of a lower desirability for individuals with higher BMI, which reduces their competitiveness in the marriage market. They demonstrate that individuals with higher BMI who marry do so at a younger age. Averett, et al. (2013) show that the direction of causality also flows in the reverse direction; that is, marriage impacts the health of individuals. The growing interest among economists in the dynamic effects of body mass on many socio-economic outcomes (e.g, employment, schooling, marriage) emphasizes the importance of understanding the effects of body mass on the accumulated determinants of wages (i.e., human capital) over the life cycle. 9
2.4 Measurement of Body Mass The body mass index is a simple means for classifying sedentary individuals by weight: underweight, ideal weight, overweight, and obese. 7 stressed its advantageous use in making population comparisons. The original proponents of this index They warned against using it for individual health diagnosis. BMI, as a function of weight and height, does not allow one to distinguish between fat mass and fat-free mass (such as muscle and bone). It may overestimate adiposity in those with more lean body mass (e.g., athletes), while underestimating adiposity on those with less lean body mass (e.g., the elderly). Different methods of assessing overweight and obesity may allow researchers to capture adiposity (or body weight) as it relates to productivity versus appearance. Fat mass can be measured by percentage of body fat (using skinfold, underwater weighing, or fat-free mass techniques) or by measures that account for mass and volume location (waist and arm circumference and the body volume index). Johansson, et al. (2009) and Wada and Tekin (2010) show that the measure of adiposity affects analysis of the obesity wage penalty. Burkhauser and Cawley (2008) find that obesity measured by BMI is only weakly correlated with obesity defined by other measures of fatness. Unfortunately, however, many data sets that contain information on employment and wages contain only measures of height and weight that can be used to construct the body mass index. Another concern is that the data on height and weight are often self-reported. These self-reports are likely to suffer from reporting error, subjectivity error, and rounding error. Because it is necessary for us to rely on the traditional BMI index to measure adiposity, we use an estimation technique that addresses potential measurement errors. Ostbye, et al. (2011) use the NLSY79 data to classify individuals into groups characterized by their BMI trajectory. These epidemiologists identify four trajectory groups: normal weight, overweight, late adulthood obesity, and early adulthood obesity. Grouping individuals together by gender, race, and cohort, they report membership rates of 35.0, 41.2, 19.7, and 4.2 percent for the four trajectory groupings, respectively. Males, blacks and those 7 Specifially, BMI equals weight in kilograms divided by height in meters squared. A BMI less than or equal to 18.5 indicates underweight; between 18.5 and 25, normal weight; greater than or equal to 25 and less than 30, overweight; and 30 or greater, obese. 10
in the younger age cohort (aged 18 to 20 years in 1981) were more likely to be in the overweight, late overweight, and early overweight trajectory groups relative to the normal weight group. In our simulations based on the estimated model, we consider lifetime BMI pathways that mimic these patterns of weight over time. 3 Description of the Research Sample The National Longitudinal Survey of Youth (NLSY) provides data on a 1979 cohort of 12686 individuals aged 14 to 22 followed annually through 1994. These individuals (and an additional cohort of adolescents that we do not include) continue to be followed every two years after 1996. Our research sample includes those who have not attrited by 1983 and who have responses for important variables in our analysis. Because the literature on the relationship between body mass and wages has detected a wage effect that differs between black and white females, we restrict our analysis in this paper to women of these races. We drop sample-eligible individuals who report an Asian race or Hispanic ethnicity (a mutually exclusive category in the NLSY data). From 1983, we follow 3213 females through 2002 or until they attrit from the NLSY survey. Table 1 displays the research sample size by year and the percent of individuals who attrit each year. 8 The research sample contains 51,884 person-year observations. The NLSY data provide a unique understanding of individual body mass dynamics over time because the survey follows the same individuals for such a long time (e.g., over 20 years in some cases) relative to most data sets. Since the evolution of body mass as individuals age is important in this study, we depict the trends in body mass over time in Figure 1. Average body mass increases by age for women of both races, with the mean increasing from 21.5 and 22.7 at age 20 to 25.9 and 29.8 at age 40, for white and black females respectively. More importantly, the 75th percentile of the BMI distribution is increasing as is the body 8 We restrict the initial 1983 sample to include individuals who are observed for at least two consecutive periods. Hence, attrition does not occur between 1983 and 1984. In fact, the 1983 data serve as initial conditions for the subsequent period of observation. Additionally, we have no need to model attrition at the end of 2002 since this is the last year of data that we use. 11
Table 1 Empirical Distribution of Research Sample Year Sample Attriters Attrition Size Rate 1983 3,213 - - 1984 3,213 67 2.09 1985 3,146 81 2.57 1986 2,065 101 3.30 1987 2,964 97 3.27 1988 2,867 52 1.81 1989 2,815 57 2.02 1990 2,758 46 1.67 1991 2,712 60 2.21 1992 2,652 35 1.32 1993 2,617 39 1.49 1994 2,578 134 5.20 1995 2,444 78 3.19 1996 2,366 102 4.31 1997 2,264 70 3.09 1998 2,194 58 2.64 1999 2,136 114 5.34 2000 2,022 42 2.08 2001 1,980 102 5.15 2002 1,878 - - Person-year observations: 51,884 12
mass of people in the right tail of the distribution as the same women age. 9 Across all ages, one-year body mass deviations average 0.21 and 0.32 BMI units for white and black females respectively. For an average height female (i.e., 5 4 ), this amounts to an increase of 1.2 to 1.9 pounds per year. Despite the fact that much of the interquartile range lies above zero at each age, the median BMI deviation is zero. Figure 1: Distribution of Body Mass as Females Age, by Race Since we intend to explore how wages vary with body mass, we emphasize that body mass increases with age, on average, in the 18 to 45 year age range. It is also the case that the strongest predictors of wages education and work experience also increase with age. In general, while additional education may be obtained at any age, it is often completed in the early twenties. Work experience, however, accumulates over the life cycle. If individuals gain weight as they age, but are also gaining work experience as they age, a depiction of 9 The horizontal lines in Figure 1 indicate upper thresholds for underweight (18.5), normal (25), and overweight (30) with a BMI greater than 30 indicating obese. The dark shaded regions indicate the interquartile range. The light shaded regions extend up to 1.5 times the interquartile range. Points on either side of the shaded areas represent remaining outliers. 13
wages by age will almost surely reveal that those with larger body mass also have higher wages. Figure 2 depicts full-time wages by age and body mass for females of different races. Visually, it appears as if wages vary by age and body mass among white females, but a less distinct pattern emerges for black women. Hence, it is important to control for human capital variables (i.e., education and work experience) that increase with age, as well as other cumulative variables that influence productivity (i.e., marital status and tenure, number of children, and body mass). Because these important wage determinants may be influenced by body mass and, inversely, influence body mass, we model their dynamic evolutions jointly while allowing for common observed and unobserved heterogeneity over the life cycle. Figure 2: Wages by Age and Body Mass Educational attainment and work experience are determined by schooling and work behaviors at each age. Figure 3 displays these probabilities by age and by predominant body mass status. 10 We model the probability of part-time employment or non-employment relative to full-time employment, and depict full-time employment probabilities by race and weight status in this figure. Marital status and fertility may change as one ages. These annual behaviors define accumulated stocks (namely, marital tenure and number of children in the household) that may shift preferences for schooling and employment. They may also impact productivity at work conditional on employment, which may be reflected in wages. Rather than focus only on births as a mechanism by which household size may change, we broaden the process to include increases through birth, adoption, and blended families and reductions through death, 10 These graphs are based on those individuals in the research sample who are observed every year from age 25 to age 40. The normal and overweight distinction reflects the body mass status of an individual in more than half of those 16 years. 14
Figure 3: Schooling and Employment Statistics by Age and Body Mass 15
marriage dissolution, removal, and matriculation. In addition to the effects children in the household may have on employment behavior and productivity, children may impact body mass of the parents. Most obviously, childbearing imposes physical changes in a woman s body mass that may not be temporary. Additionally, children impose both time and financial requirements that may alter caloric intake and expenditure of their caregivers. Figure 4 describes marriage and child accumulation probabilities for females by predominant body mass status. Figure 4: Marriage and Children Statistics by Age and Body Mass 16
4 Empirical Framework Our empirical framework is motivated by a theoretical model of life cycle decisionmaking and health production: per-period decisions regarding education, employment, marriage, and children with uncertain future wages and body mass transitions. The theory allows us to derive choice probabilities that we approximate using dynamic, non-linear functions of the information available (to an individual) at the beginning of each decisionmaking period. The theory also describes how these period-by-period decisions affect the production of health (or evolution of body mass) over time and the implied distribution of wages. We then estimate the approximated demand and production equations jointly, and allow for both permanent and time-varying individual unobserved heterogeneity (UH) through the use of discrete factor random effects (Heckman and Singer, 1983 and Mroz, 1999). Rather than estimate the conditional mean of wages and body mass, we estimate the conditional density of these observed outcomes using a conditional density estimation (CDE) technique (Gilleskie and Mroz, 2004). This multi-faceted empirical approach enables us to consider aspects of the problem that may not be computationally-feasible with full-solution and estimation of the dynamic optimization problem. We 1.) include several endogenous explanatory variables that provide a detailed description of the histories of individual decisions over the life cycle, 2.) consider the continuous evolution of body mass rather than discrete categories as several authors have done, and 3.) explore the possibility that variables of interest may have different marginal effects at different points of support of the wage and body mass distributions. We seek to evaluate the causal pathways of body mass on wages of women (either directly or indirectly through other life cycle determinants) and to distinguish the contemporaneous effect from that generated by life cycle behaviors influenced by body mass. 17
4.1 Timing and Notation In each year t after age 18, an individual obtains a wage offer drawn from the population distribution of wages. The individual also observes her spouse s wage if married. 11 The wage a female would receive if she worked, w t, depends on education and employment experience, as described by Mincer (1974). We argue that these human capital variables may not fully capture a person s productivity at work. Her productivity, or effort, might be influenced by her health or her time demands. Thus, we include body mass (Cawley, 2004), marital status (Korenman and Neumark, 1991), and number of children entering period t as determinants of the wage distribution. 12 Each period the individual then jointly decides 1.) whether or not to attend school, 2.) whether to be non-employed, part-time employed, or full-time employed, 3.) whether or not to be married, and 4.) whether or not to acquire (or lose) a child (or children) living in the household. We let indicator variables (lower case) define the alternatives. That is, s t = 1 if the individual attends school in period t and s t = 0, otherwise. The employment indicator takes on the value e t = 0 if the individual does not work in period t, e t = 1 if he works part time (less than 1375 hours per year), and e t = 2 if he works full time (1375 hours or more per year). We indicate the marriage decision by m t = 1 if the individual is married in year t, and m t = 0, otherwise. The number of children in the household may increase or decrease due to pregnancy (singleton or multiples), marriage, divorce, age of child, or child mortality. 13 The variable k t = 0 indicates that no children are acquired in year t, and values of k t = 1 and k t = 1 indicate that at least one child is acquired or lost in year t. 11 Spouse income is treated as exogenous. While we model the marriage decision, which may be influenced by body mass, we do not model the correlation between an individual s body mass and her potential spouse s income (Chiappori, et al., 2012). 12 We recognize that body mass is an imperfect measure of health, and may not fully capture the role that health plays in one s productivity. Instead, or additionally, it may signal expected medical care expenditures (which may be compensated by the employer and hence affect offered wages) or reflect an employer s tastes based on appearance. We cannot distinguish these roles in estimation. 13 A reason for a change in number of children is available in the data. For example, the survey records whether a child was born, died, adopted, etc. Additionally, the a comparison of the number of children in the household from year to year reveals additions and losses of more than one child in a non-trivial number of cases. The changes could be due to birth by the sample female as well as children acquired (or lost) through other channels (e.g., adoption, marriage, divorce, death). Because of the large number of ways household size can change, we are motivated to focus only on the change in number of children in the household rather than the channel. A larger household size, regardless of channel, places demands on time, finances, and energy which may impact per-period decisionmaking. 18
These yearly decisions (i.e., school attendance, employment, marriage, and children) produce stock variables upon entering period t that summarize the history of the decisions. The vectors of these history variables (upper case) are denoted: educational history (S t ), work experience (E t ), marital history (M t ), and household child accumulation (K t ). These vectors include both duration values as well as indicators of behavior or outcomes in the last period (t 1) and polynomials of the continuous values. These histories up to period t influence the value of each period t schooling, employment, marriage, and child alternative. 14 We allow the vector of variables that capture body mass history up to the beginning of a period (B t ) to affect current schooling, employment, marital status and child accumulation decisions. Other information known at the beginning of year t that may influence preferences, constraints, or expectations includes exogenous individual characteristics (e.g., age, gender, race, urbanicity, region) and exogenous aggregate business cycle indicators (e.g., calendar year trend variables) denoted by the vector X t. Variables that capture known price and supply conditions such as average tuition, wage and employment information, welfare amounts, restaurant sales, and costs of different food items are denoted by the vector P t. We use unique superscripts to denote, for descriptive purposes only, the variables in P t that are related to particular endogenous behaviors and outcomes; that is, P t = (P S t, P E t, Pt M, P K t, P B t ). We recognize, however, that the optimal demand functions for each behavior may depend on own and cross prices. The vector of individual information known at the beginning of period t is denoted Ω t = (B t, S t, E t, M t, K t, X t, P t ). As forward-looking decisionmakers, the individuals face uncertainty about future wages. Wages depend on educational attainment, work experience, and marital and child histories. In addition to these human capital variables, wages may also be influenced by body mass, which explains variation in wages caused by health impacts on productivity or employer preferences about appearance. Future body mass is also uncertain. Having made the beginning-of-period decisions (i.e., education, employment, marriage, and child accumulation) that affect the per-period budget and time constraints, we assume that individuals then 14 We are not explicit about how these variables may impact preferences or the budget constraint because we do not estimate the primitive parameters of the decisionmaking optimization problem. However, these variables, because they are a part of the information set when evaluating alternatives, are arguments of the non-linear demand functions resulting from optimization. 19
allocate income and time to the daily activities of caloric intake and caloric expenditure (i.e., eating and exercising) denoted ci t and ce t. Theoretically, the density that describes possible values of next period s body mass, B t+1, depends on body mass this period and these daily caloric input and output behaviors during the period. Unfortunately, these latter activities are not observed in our data. The daily input decisions depend on market and time prices of food and exercise, P B t, as well as the less frequently changed behaviors regarding education, employment, marital, and child accumulation. Substituting in the determinants of caloric intake and expenditure, expectations of body mass transition are, indirectly, a function of the period t behaviors, the resulting income (own wages and spouse s wages if married), and relevant prices (P B t ). The timeline below depicts the observed and unobserved outcomes per period (e.g., one year), and the exogenous and endogenous information available at the beginning of each period. beginning of t unobserved wage draw s t, e t,m t, k t w t unobserved demand for caloric intake and caloric expenditure B t+1 beginning of t + 1 w t observed demand for schooling, employment, marriage, and family size observed wage if employed ci t, ce t observed body mass evolution Ω t = (B t, S t, E t, M t, K t, X t, P t ) Ω t+1 = (B t+1, S t+1, E t+1, M t+1, K 1t+1, X t+1, P t+1 ) Variables that explain the observed outcomes are described in Tables 2, 3 and 4 for our research sample. The endogenous variables are summarized over all person-observations in the sample over the 1984-2002 period (Table 2). Since our focus is on the role of body mass, we discuss here the variables we use in estimation to capture the history of one s body mass. We use moments of the continuous index of body mass (BMI) rather than categories of adiposity (i.e., underweight, normal weight, overweight, and obese). We report both for statistical purposes. We also use time-varying indicators of ever overweight or ever obese to capture the contemporaneous effects of past weight issues on current behaviors and 20
wages conditional on current BMI. Finally, we include a variable representing how different a person s BMI is from women of the same race in each year. This measure could reflect social capital, as deviations from the norm in body mass among a specific group might reflect perceptions of effort or commitment. The exogenous time-invariant individual variables are summarized for the 3212 females in our research sample in 1984, while the time-varying variables are summarized over all person-years (Table 3). Own citizenship and parental information are used to identify initially-observed values of endogenous variables (discussed in Section 4.6). The exogenous price and supply-side variables vary by state or county and time, and are described in Table 4. While there are several sources of statistical identification in our model, these exogenous time-varying variables serve as classic identification variables since they are included in theoretically-relevant demand (behavior) equations and excluded from the wage and body mass production function conditional on included endogenous behaviors (as described in more detail in the next section). 15 15 We obtain variation at the state, county (if available), and year levels. Average tuition rates are from the Higher Education General Information Survey (www.icpsr.umich.edu/icpsrweb/icpsr/series/30) and vary by state and year. Employment data and retail, food, and restaurant sales data are from the Woods and Poole Economics Database (www.woodsandpoole.com) and vary by state and county and year. We use state and year level variation in the average Aid to Families with Dependent Children (AFDC) payment (Bernal and Keane, 2010). Yearly population data are from the Census Bureau (www.census.gov/popest/data/historical) with total population at the county level and sex ratios at the state level. Food, alcohol, and cigarette prices, from the Council for Community and Economic Research (formerly the American Chamber of Commerce Research Association or ACCRA), are aggregated to the county level by year (www.c2er.org). The food price consists of weighted prices of frying chicken (per pound), chunk light tuna (5 ou.), half gallon of whole milk, a dozen eggs, margarine (1 lb.), grated parmesan cheese (8 ou.), potatoes (10 lb.), bananas (per pound), iceberg lettuce, and white bread. The junk food price is the average price of a McDonald s quarter pound cheeseburger, a thin crusted cheese pizza at Pizza Hut or Pizza Inn, and fried chicken at KFC or Church s. 21
Table 3 Description of Endogenous Individual Explanatory Variables White (N=1951) Black (N=1262) Variable name Mean Std Dev Mean Std Dev Endogenous individual variables over all person-years Body mass B t BMI in t 23.92 5.06 26.56 6.10 Underweight: BMI in t 18.5 0.05 0.22 0.03 0.16 Normal weight: 18.5 BMI in t < 25 0.65 0.48 0.46 0.50 Overweight: 25 BMI in t < 30 0.19 0.40 0.27 0.44 Obese: BMI in t 30 0.11 0.31 0.24 0.43 Ever overweight prior to t 0.39 0.49 0.57 0.50 Ever obese prior to t 0.14 0.35 0.28 0.45 Weight in pounds at t 142.27 31.14 155.88 36.44 Height in inches at t 64.66 2.59 64.26 2.90 Standardized deviations from mean BMI in t 0.00 1.00 0.00 1.00 Education history S t Enrolled in t-1 0.09 0.29 0.08 0.27 Years enrolled in school entering t 13.23 1.95 12.70 1.73 Years enrolled < 12 entering t 0.09 0.28 0.17 0.38 Years enrolled 12 entering t 0.91 0.28 0.83 0.38 Years enrolled 16 entering t 0.15 0.36 0.07 0.26 Freshmen year of college in t 0.02 0.13 0.02 0.13 Employment history E t Employed in t-1 0.85 0.36 0.79 0.41 Employed full time in t-1 0.69 0.46 0.71 0.45 Employed part time in t-1 0.31 0.46 0.29 0.45 Years employed entering t 10.34 5.47 8.94 5.80 Years full time employed entering t 6.42 5.15 5.43 5.31 Years part time employed entering t 3.92 3.00 3.51 2.70 Marital history M t Married in t-1 0.63 0.48 0.31 0.46 Years married entering t if married in t-1 8.00 5.21 6.64 4.87 Years newly single entering t if single in t-1 1.64 3.15 1.67 3.56 Child history K t Number of children entering t 1.12 1.16 1.42 1.29 Acquire any children in t-1 0.09 0.28 0.09 0.28 Lose any children in t-1 0.02 0.12 0.03 0.17 22
Table 2 Description of Exogenous Individual and Aggregate Explanatory Variables White (N=1951) Black (N=1262) Variable name Mean Std Dev Mean Std Dev Time-invariant individual variables in year 1984 AFQT score minus median 19.23 25.47-11.57 19.81 AFQT score missing 0.03 0.16 0.02 0.12 Non-US citizenship at birth 0.03 0.15 0.03 0.17 Non-US citizenship at age 14 0.01 0.07 0.01 0.10 Mother is non-us citizen 0.04 0.20 0.03 0.17 Father is non-us citizen 0.14 0.35 0.03 0.16 Years of education of mother 12.06 2.38 10.81 2.55 Mom s education missing 0.03 0.18 0.07 0.25 Years of education of father 12.39 3.16 10.23 3.52 Dad s education missing 0.08 0.27 0.28 0.45 Time-varying individual variables over all person-years Age in years - 18 12.28 6.11 12.19 6.08 Rural residence 0.26 0.44 0.17 0.37 Residence type missing 0.05 0.23 0.04 0.20 Northeast region 0.20 0.40 0.14 0.35 North central region 0.34 0.47 0.18 0.39 West region 0.16 0.37 0.07 0.25 South region (omitted category) 0.30 0.46 0.60 0.49 State of residence missing 0.01 0.08 0.01 0.10 Trend and Interval Variables Mean Std Dev Min Max Time trend (1=1984) 9.16 5.40 1 19 Continuous interval measure -1.42 0.68-2.20 0.00 23
Table 4 Description of Exogenous Price and Supply-Side Variables Variable name Mean Std Dev Min Max Schooling variables Pt S Two year college semester tuition in 000s 1.483 0.842 0.127 5.027 Four year college semester tuition in 000s 2.338 0.986 0.339 6.868 Graduate school semester tuition in 000s 2.644 1.139 0.518 7.076 Employment variables Pt E Unemployment rate 6.86 2.26 1.90 17.70 Total employment per capita 0.58 0.11 0.38 1.34 Ratio of manuf empl to total empl 0.13 0.05 0.02 0.26 Ratio of service empl to total empl 0.28 0.05 0.18 0.48 Total earnings per employee 41.18 8.08 27.41 77.16 Ratio of manuf earnings to total earnings 0.18 0.08 0.02 0.39 Ratio of service earnings to total earnings 0.24 0.05 0.13 0.42 Marriage and Children variables Pt M and Pt K Total population in 000,000s 51.07 56.35 4.54 350.25 White gender ratio: male, 20-60/female, 15-50 1.05 0.07 0.92 1.64 Black gender ratio: male, 20-60/female, 15-50 0.89 0.09 0.75 1.79 Household income in 000s 59.62 12.29 36.97 106.29 AFDC per month per family of four in 00s 5.23 1.96 1.47 11.87 Body Mass variables Pt B Mean price of food 1.85 0.15 1.51 2.76 Mean price of junk food 4.69 0.32 3.57 6.79 Mean price of carton of cigarettes 19.59 6.67 10.03 47.64 Mean price of 6-pack of beer 4.87 1.04 3.46 8.18 Mean price of bottle of wine 6.28 0.90 3.93 10.47 Mean price of liter of liquor 17.41 4.25 8.63 26.20 Ratio of food sales to total retail sales 0.18 0.03 0.08 0.25 Ratio of restaurant sales to total retail sales 0.10 0.03 0.06 0.32 Indicator of missing price data 0.06 0.24 0.00 1.00 Note: Data presented are means over 50 states and the District of Columbia for the years 1984-2002. Dollar amounts are in year 2000 dollars. 24